Skip to content

Planning for Randomizers

Last updated on 21/02/2024

I recently took a look at how we can prove a given instance of a randomizer romhack can be won using propositional logic. It’s posted over on the company’s blog, so make sure to take a look at it over there.

It’s kind of part of the same thread as the “Five everyday AI Problems piece” I have sitting in the hopper over there. The notion being that we all encounter problems that come up in AI textbooks every day without recognizing them. Here are some examples so that you know you’re looking at an AI problem the next time you encounter one.

Beyond that though, it’s the start of really hammering on bringing some AI techniques to bear on that particular problem. The first sort of glances in the direction of propositional logic and says “Yeah, you can encode that problem this way and solve it.” I’d like to go much deeper and with more rigor. Specifically, I’d like to:

  • encode the problem in PDDL and get fast downward to solve some instances.
  • do routing to determine shortest tours of bosses and treasure chests.
  • encode the problem as an MDP and produce a next suggested action tracker

We’ll see if and when I actually get around to those things, but that’s what I have in mind.

Published inArtificial Intelligence